Two open-weight models joined AWS's catalog
AWS announced access to OpenAI's gpt-oss-120b and gpt-oss-20b through Amazon Bedrock and Amazon SageMaker AI in August 2025. The announcement concerned those two open-weight models, not every model sold through OpenAI's own platform or a transfer of the ChatGPT service to AWS.
Amazon described managed access through Bedrock and additional deployment and customization tools through SageMaker AI. It also identified some capabilities as future additions. Those roadmap items should not be counted as universally available on the original launch day.
OpenAI's release allowed separate deployment
OpenAI released the model weights under Apache 2.0 and described gpt-oss as its first open-weight language-model release since GPT-2. The company's introduction distinguished models that developers could run and customize from its hosted API offerings. Making weights available did not turn every other OpenAI product into an open-weight release.
The release also described different hardware requirements for the two sizes and adjustable reasoning effort. Those were the developer's documented characteristics and claims, not an independent performance evaluation by this publication. A deployment still needs an appropriate runtime and resources for its actual workload.
Model choice and service choice are separate
A model's weights define one part of an application; the service around them determines how it is operated, monitored and connected to other systems. The AWS announcement expanded a deployment option without making all providers' features interchangeable. For this historical release, the important distinction is between availability of specific open-weight models and a broader commercial partnership or hosted product. Neither an open license nor a cloud listing alone establishes the cost, security or suitability of a particular application.